Document subjectivity and target detection in opinion mining using HMM POS-Tagger

With the abundance of digital text documents, opinion mining is an emerging research topic at this time. A significant problem in opinion mining is a problem of document subjectivity. This problem is related to how to determine a sentence that is extracted from a text is an opinion or not. Another important issue is to find the target of an opinion. In the business world opinion mining is used for automatically analyzing customer opinions about products. In the education world opinion mining can be applied to evaluate and utilize the textual advices in an educational evaluation using questionnaire. In this study we used a technique of two order HMM Based Part-of-Speech (POS) Tagger to identify opinions from the text. This technique was also used to determine the target opinion in an opinion sentence. Both document subjectivity and target opinion was implemented using rule-based techniques, i.e the use of patterns of part-of-speech couples as a characteristic pattern of opinion and target opinion. The data used for the evaluation was student comments of learning evaluation in IST AKPRIND Yogyakarta. The results showed that the precision and recall of subjectivity documents amounted to 0.95 and 0.92, and for the detection targets are 0.91 and 0.89.